Biology subjects

Pisula, J. I.

Publications and source records attributed to Pisula, J. I..

2 recordsLinked to original sources

bioRxiv2024

Deep learning-based interpretable prediction of recurrence of diffuse large B-cell lymphoma

BackgroundThe heterogeneous and aggressive nature of diffuse large B-cell lymphoma (DLBCL) presents significant treatment challenges as up to 50% of patients experience recurrence of disease after chemotherapy. Upfront detection of recurring patients could offer alternative treatments. Deep learning has shown potential in predicting recurrence of various cancer types but suffers from lack of interpretability. Particularly in prediction of recurrence, an understanding of the models decision could eventually result in novel treatments. MethodsWe developed a deep learning-based pipeline to predict recurrence of DLBCL based on histological images of a publicly available cohort. We utilized attention-based classification to highlight areas within the images that were of high relevance for the models classification. Subsequently, we segmented the nuclei within these areas, calculated morphological features, and statistically analyzed them to find differences between recurred and non-recurred patients. ResultsWe achieved an f1 score of 0.83 indicating that our model can distinguish non-recurred from recurred patients. Additionally, we found that features that are the most predictive of recurrence include large and irregularly shaped tumor cell nuclei. DiscussionOur work underlines the value of histological images in predicting treatment outcomes and enhances our understanding of complex biological processes in aggressive, heterogeneous cancers like DLBCL.

bioinformatics↗

bioRxiv2022

Predicting the HER2 status in esophageal cancer from tissue microarrays using convolutional neural networks

BackgroundFast and accurate diagnostics are key for personalized medicine. Particularly in cancer, precise diagnosis is a prerequisite for targeted therapies which can prolong lives. In this work we focus on the automatic identification of gastroesophageal adenocarcinoma (GEA) patients that qualify for a personalized therapy targeting epidermal growth factor receptor 2 (HER2). We present a deep learning method for scoring microscopy images of GEA for the presence of HER2 overexpression. MethodsOur method is based on convolutional neural networks (CNNs) trained on a rich dataset of 1,602 patient samples and tested on an independent set of 307 patient samples. We incorporated an attention mechanism in the CNN architecture to identify the tissue regions in these patient cases which the network has detected as important for the prediction outcome. Our solution allows for direct automated detection of HER2 in immunohistochemistry-stained tissue slides without the need for manual assessment and additional costly in situ hybridization (ISH) tests. ResultsWe show accuracy of 0.94, precision of 0.97, and recall of 0.95. Importantly, our approach offers accurate predictions in cases that pathologists cannot resolve, requiring additional ISH testing. We confirmed our findings in an independent dataset collected in a different clinical center. ConclusionsWe demonstrate that our approach not only automates an important diagnostic process for GEA patients but also paves the way for the discovery of new morphological features that were previously unknown for GEA pathology.

bioinformatics↗